Generative artificial intelligence tools can make studying feel considerably easier, but this subjective sense of ease may distort how accurately students assess their actual mastery of academic material. A research investigation conducted by Juzhi Zhang at University College London and published in the journal Frontiers in Psychology evaluated how AI assistance, explanation fluency, and subsequent unaided retrieval influence subjective judgments of learning relative to objective performance on independent tests.
Grounded in the cue-utilization framework, the project investigated processing fluency as a key metacognitive mechanism connecting generative artificial intelligence assistance with learners’ judgments of learning and metacognitive calibration. Across a series of three controlled experimental studies with university students, the researcher examined whether the cognitive ease experienced during AI-assisted study sessions genuinely reflects learning gains or primarily alters students’ internal appraisals of their knowledge.
In the first experiment, which enrolled two hundred sixty-four university students, researchers tested the direct and indirect consequences of utilizing generative artificial intelligence during learning tasks. The findings demonstrated that generative artificial intelligence assistance significantly increased both subjective processing fluency and participants’ judgments of learning. Furthermore, statistical evaluation identified a significant indirect association linking artificial intelligence assistance to heightened judgments of learning specifically through increased processing fluency.
To evaluate the specific attributes of artificial intelligence output, a second study involving three hundred sixty university students manipulated fluency across scripted explanations delivered through a simulated generative artificial intelligence interface, comparing these conditions against conventional text-based learning materials. Participants receiving high-fluency explanations reported elevated judgments of learning and exhibited greater positive calibration bias compared with participants assigned to low-fluency explanations or traditional text formats.
Despite the pronounced increases in perceived learning and calibration bias observed in the high-fluency condition of the second study, statistical analyses did not detect statistically significant between-condition differences in objective test performance. In addition, absolute calibration error did not differ significantly across the experimental conditions, indicating that while students felt more confident when reading fluent artificial intelligence responses, their immediate demonstrable comprehension did not experience a measurable corresponding advantage.
The third study, encompassing three hundred eighty-six university student participants, examined whether engaging in an unassisted recall exercise could recalibrate learners’ subjective evaluations. The data revealed that completing a brief, unaided retrieval attempt without artificial intelligence assistance substantially weakened the relationship between processing fluency and judgments of learning. This intervening retrieval task also succeeded in reducing overall positive calibration bias among the student participants.
While the brief retrieval practice effectively lowered overall positive calibration bias in the third study, the magnitude of this reduction did not differ significantly between the high-fluency and low-fluency experimental conditions. The retrieval intervention demonstrated consistent effects across both levels of explanation fluency, indicating that attempting recall independently helped temper inflated confidence regardless of how smoothly the initial instructional material had been presented by the system.
Taken together, the findings across the three experiments indicate that generative artificial intelligence can systematically alter metacognitive judgments by shifting how fluent learning feels to students, even though experimental analyses detected no statistically significant between-condition differences in immediate independent test performance. Providing students with structured opportunities for unaided retrieval may help redirect their judgments away from artificial intelligence-generated processing ease toward diagnostic evidence of what they can independently retrieve.
Subject of Research: Metacognitive calibration and processing fluency in generative artificial intelligence learning environments
Article Title: When AI makes learning feel effortless: how processing fluency shapes students’ metacognitive judgments
Article References: Zhang, J. (2026). When AI makes learning feel effortless: how processing fluency shapes students’ metacognitive judgments. Frontiers in Psychology, 17, Article 2003031. https://doi.org/10.3389/fpsyg.2026.2003031
Image Credits: AI Generated
DOI: 10.3389/fpsyg.2026.2003031
Keywords: generative artificial intelligence, processing fluency, judgments of learning, metacognitive calibration, retrieval practice, metacognition, educational psychology, university students, calibration bias, independent test performance, unaided retrieval, cue-utilization framework, subjective learning judgments, explanation fluency, simulated AI interface, conventional text learning, absolute calibration error
Cite Scienmag News
Glenn Wilkins. (November 22, 2026). AI Explanations Can Inflate Student Confidence Without Improving Test Performance. Scienmag. https://scienmag.com/ai-explanations-can-inflate-student-confidence-without-improving-test-performance/
Glenn Wilkins. "AI Explanations Can Inflate Student Confidence Without Improving Test Performance." Scienmag, 22 November 2026, https://scienmag.com/ai-explanations-can-inflate-student-confidence-without-improving-test-performance/. Accessed 22 November 2026.
Glenn Wilkins. "AI Explanations Can Inflate Student Confidence Without Improving Test Performance." Scienmag. November 22, 2026. https://scienmag.com/ai-explanations-can-inflate-student-confidence-without-improving-test-performance/








